Skip to content

AI animation and motion capture tools for indie games (2026)

An indie studio can now get usable body motion from a phone video or a text prompt for between nothing and about 20 dollars a month, exported as FBX or BVH. What you get is a first pass: fingers, feet and anything hidden from the camera still need an animator's time. Free tiers are mostly for testing, because several services allow commercial use only on a paid plan, and Tencent's open text-to-motion model cannot be used in the EU, the UK or South Korea.

By Futigo · Updated · 9 min read

Key points

  • Rokoko Vision, Plask, QuickMagic, DeepMotion and Move One all turn ordinary video into skeletal animation, with free allowances from 15 seconds a day to 60 seconds a month.
  • DeepMotion and QuickMagic state that their free plans are not for commercial use. Uthana and Cascadeur also tie commercial use to payment.
  • NVIDIA's Kimodo is the open text-to-motion model with the friendliest terms: Apache 2.0 code and weights that NVIDIA's licence calls commercially usable.
  • Godot and Unity both retarget humanoid animation in the engine. We found no retargeting tool in Blender's own manual or its 5.2 release notes, so that step uses an add-on.
  • For 2D games, sprite-sheet generators such as Ludo.ai and PixelLab are the relevant tools, and generated frames still need checking for consistency one by one.

Hosted tools side by side

These nine were live with a public pricing page, FAQ or documentation we could read on 15 November 2026. The first five are markerless motion capture, Uthana and QuickMagic also generate motion from text, Cascadeur is a desktop animation programme with AI assistance, and the last two make 2D sprite animation.

Hosted AI animation and motion capture tools compared: function, input, export formats, free tier, entry paid plan and commercial terms, November 2026
ToolWhat it doesInputExport formatsFree tierEntry paid planCommercial use
Rokoko Vision 3.0Single-camera video to body motionUploaded video, one performerFBX; BVH on paid plansStarter: 30 seconds of processing a monthBasic: $12 a month, or $10 a month billed annually, 600 secondsNot stated for Vision output on the pricing page
Move One (Move AI)Single-camera video to body motionVideo, clips up to 30 seconds free and 60 seconds paidNot confirmed30 one-time creditsStarter: $18 a month excluding VAT, 60 creditsNot confirmed
DeepMotion Animate 3DVideo to motion with optional face and hand trackingMP4, MOV or AVI videoFBX, BVH, MP4; GLB for custom charactersFreemium: 60 seconds a monthSee pricing pagePaid subscription required; free plan is personal and non-commercial
PlaskVideo to body and hand motion, multi-personVideoFBX, GLB, BVH15 seconds a day, 1 GB storageStandard: $18 a month billed annually, 10 minutes a monthNot stated on the pricing page
QuickMagicVideo to motion, and text to motionVideo up to 30 seconds free, or a text promptFBX on free; paid adds BVH, Unity Anim, Unreal, Mixamo and others50 V Coins, lowest queue priority, assets kept 15 daysBasic: $11.90 a month billed yearly, 200 creditsPaid plans only
UthanaText to motion, video to motion, auto-riggingText prompt, video, or a character to rigNot confirmedNone listedPay as you go: text to motion $0.02 to $0.10 a second, video to motion $0.05 to $0.06 a second, $0.15 a rigIncluded with paid usage
CascadeurKeyframe animation with AI posing, AI inbetweening and physics toolsYour rigged characterFBX, DAE, USD on paid plans; its own format only on FreeFree plan, non-commercial, no FBX exportIndie: €19 a month, or €8 a month billed annuallyIndie up to $100,000 annual revenue; Pro unlimited
Ludo.aiAnimated 2D sprite sheetsText prompt, keyframes or a reference videoTransparent PNG sprite sheets30 credits at sign-upIndie: $15 a month billed yearly, 3,000 credits a yearLudo says you own output on every plan
PixelLabPixel-art animation from a skeleton or text, 4 and 8 direction rotationsSprite, text prompt, skeleton posesNot confirmedFree trial, no card requiredSee pricing pageNot confirmed

Read from each tool's own pricing page, FAQ or documentation on 15 November 2026, as of November 2026. Rokoko's page showed a second, higher set of prices lower down; we used its main table. QuickMagic's price is its discounted yearly rate against a listed $14.90. DeepMotion's and PixelLab's plan prices did not load. Ludo's free credits and ownership statement come from a comparison page on Ludo's own site.

Units differ. Move One charges one credit per person per second on its first-generation model and two on its second, so 60 credits is 30 to 60 seconds of motion a month. DeepMotion charges one credit a second, plus half a credit each for face and hand tracking. Uthana ended its subscriptions in August 2026 and now sells credits that do not expire.

Two names have changed. Rokoko's Vision 3.0 accepts uploaded video only: live webcam capture has gone and dual-camera capture is being retired. Wonder Studio, which also extracts motion from footage, is now Autodesk Flow Studio; its pricing page would not load for us, so it is not in the table.

Open models and free tools you can run yourself

Running a model locally removes the per-second charge and keeps your footage on your machine. You need an NVIDIA GPU for most of these, and the licences vary more than the hosted terms do.

Open motion models and free animation tools compared: developer, licence, commercial use, output, stated hardware and GitHub stars, November 2026
NameDeveloperLicence and commercial useWhat it generatesStated hardwareGitHub stars
KimodoNVIDIACode Apache 2.0. SOMA and G1 weights under the NVIDIA Open Model License, which says models are commercially usable. The SMPL-X weights are under a research licenceMotion from text plus keyframe, path and waypoint constraints. NPZ, and BVH for the SOMA skeletonAbout 17 GB VRAM, or under 3 GB with the text encoder on the CPU4,220
HY-Motion 1.0TencentTencent HY-Motion 1.0 Community License. Commercial use allowed under 1 million monthly active users, not in the EU, UK or South KoreaSkeleton-based human motion from text26 GB VRAM, or 24 GB for the Lite model2,671
FlowHMRFlowHMR authorsCode Apache 2.0. Weights under the licence on their model card; some components need separate licence acceptanceMotion from video, saved as NPZ and optionally FBX12 GB VRAM or more669
GVHMRZhejiang UniversityCustom licence: education, research and non-profit only. For commercial use the licence says to contact the authorsMotion from videoNot stated here2,296
FreeMoCapFreeMoCap ProjectAGPL 3.0. Commercial use allowed under copyleft terms; other terms on requestCamera-based motion capture system, described as research-gradeNot stated here10,816
UniRigTsinghua University and TripoMIT. Commercial use allowedSkeleton and skin weights for a meshNot stated here2,130
SkinTokensVAST (Tripo)MIT. Commercial use allowedRigging, the successor to UniRigNot stated here646
ToonCrafterResearch project, SIGGRAPH Asia 2024Code Apache 2.0In-between frames for cartoon images, up to 16 frames at 512 by 320About 24 to 27 GB VRAM; community builds about 10 to 12 GB6,062

Licences read from each repository's licence file and README; stars from the GitHub API. All checked on 15 November 2026. Research code often depends on body models and datasets with their own, stricter terms, so read the README before shipping output.

Kimodo, first released in March 2026 and updated in October, was trained on 700 hours of motion capture that NVIDIA describes as commercially friendly. HY-Motion 1.0 lists its own limits plainly: humans only, no multi-person interaction, no looping or in-place motion. Its licence bars using the output outside its territory, which rules out a game sold worldwide. FreeMoCap is under active development, with a 2.0 alpha released on 11 November 2026.

How markerless capture works, and where it goes wrong

A studio system tracks reflective markers from many cameras. A markerless tool has one view. A neural network estimates where each joint is in every frame, then a solver fits a skeleton to those estimates and guesses depth, global position and contact with the floor. Every fault follows from how much has to be guessed.

  • Occlusion. Rokoko's own page says the solver needs a complete view of the performer. An arm behind the body or a leg behind a chair is invented, not measured.
  • Foot sliding. With no floor sensor, planted feet drift. DeepMotion offers foot locking, physics simulation and motion smoothing as settings, which tells you what the raw result needs.
  • Hands and fingers. Rokoko Vision had no finger tracking at launch and points to its gloves. DeepMotion and Plask track hands, DeepMotion at extra credit cost.
  • One performer. Rokoko captures a single person per recording. DeepMotion says it can track up to eight from one video.
  • Footage quality. Background, lighting and camera distance all change the result. Film the whole body, in even light, against an uncluttered background.

Retargeting onto your own skeleton

Captured motion arrives on the tool's skeleton, not yours. Retargeting maps it across. For a model that has no skeleton yet, Uthana rigs one for $0.15, UniRig does it locally, and our AI 3D model generator comparison covers the rigging options in Meshy and Tripo.

  • Godot. The advanced import settings hold a Retarget section on the Skeleton3D node. You assign a BoneMap with the SkeletonProfileHumanoid preset, which maps bones automatically by name, then use Rest Fixer to match the reference T-pose. Its Normalize Position Tracks option exists to reduce foot slipping between models of different heights.
  • Unity. The Unity 6.6 manual requires each model to use the Humanoid animation type with a configured Avatar, set up in a T-pose. Any humanoid clip then plays on any humanoid model.
  • Blender. Blender's manual documents BVH import, but we found no retargeting tool there or in the 5.2 animation release notes. Add-ons fill the gap, for example the GPL-licensed Retarget extension for Blender 5, which has presets for Mixamo, Unreal and other rigs.

All three assume a humanoid. Creatures, quadrupeds and props with odd skeletons need custom bone maps or hand animation.

2D games: frames and cut-out rigs

Most 2D games do not use motion capture at all. They use frame animation, where every frame is a separate drawing, or cut-out animation, where one set of parts is moved and rotated. Godot's cut-out animation guide notes you can mix the two, keeping frames for hard parts such as hands and faces.

  • Sprite-sheet generators. Ludo.ai makes animated sheets from a prompt, keyframes or a reference video, at 1.5 to 20 credits an animation. PixelLab lists skeleton animation, text animation and directional rotation for pixel art, with an Aseprite extension.
  • In-betweening. ToonCrafter interpolates between two cartoon frames, but only up to 16 frames at 512 by 320, and its README calls it a research exploration, not a product. PixelLab's documentation marks its own interpolation tool as old.
  • Gone. Meta's Animated Drawings, which animated a drawn figure from a single image, is archived on GitHub and no longer maintained.

The weak point for generated frames is consistency: a character's proportions, palette and outline have to match across every frame and every animation. A cut-out rig avoids that by construction, since the parts never change.

Free motion libraries that are not AI

Before capturing a walk cycle, check whether someone has already given one away. Quaternius's Universal Animation Library has more than 120 animations on a humanoid rig made for retargeting in Godot, Unity and Unreal, under CC0, in FBX and glTF. Rokoko's pricing page says animations from its Motion Library can be used in any project, including commercial ones.

Adobe's Mixamo is the best-known free library and auto-rigger for humanoids. Adobe's FAQ page would not load for us today, so we have not restated its licence here. Read it before shipping.

Clean-up time is the real cost, and what we do at Futigo

The subscription is the small number. A captured clip still has to be trimmed, looped, retargeted, have its feet pinned and its hands posed, and be checked on the real character at game camera distance. Text-to-motion skips the filming but not the rest, and HY-Motion's own notes say it cannot make looping or in-place motion, which is what most game locomotion is. Budget animator hours per clip, not credits.

  • Good fit: background characters, crowd motion, blocking and prototypes, reference for hand keying.
  • Poor fit: hero animation close to the camera, finger work, two characters in contact, creatures, tightly timed combat.

Our games are mostly 2D, built in Godot 4.7 with much of the art drawn in code, so this article is desk research from pricing pages, documentation and licence files, not a hands-on benchmark. For engine choice, see our open-source game engine comparison.

Questions

Can I do motion capture with just a phone?
Yes. Rokoko Vision, Move One, DeepMotion Animate 3D, Plask and QuickMagic all estimate body motion from a single ordinary video. Expect to fix foot sliding, hands and any limb the camera could not see.
What is the best free AI motion capture tool?
For testing, Plask's free plan gives 15 seconds a day with FBX, GLB and BVH export, and DeepMotion gives 60 seconds a month. DeepMotion and QuickMagic say their free plans are not for commercial use, so check the terms before shipping anything made on a free tier.
Is there an open-source text-to-motion model I can use commercially?
NVIDIA's Kimodo has Apache 2.0 code and SOMA-skeleton weights under the NVIDIA Open Model License, which states the models are commercially usable. Tencent's HY-Motion 1.0 also allows commercial use under 1 million monthly active users, but not in the EU, the UK or South Korea.
How much does AI motion capture cost in 2026?
Entry paid plans run from about $10 to $18 a month as of November 2026: Rokoko Basic is $10 a month billed annually for 600 seconds, QuickMagic Basic is $11.90 billed yearly, and Plask Standard and Move One Starter are $18. Uthana charges per second instead, from $0.05 for video to motion.
Can Godot retarget motion capture animation?
Yes. In the advanced import settings, assign a BoneMap with the SkeletonProfileHumanoid preset to the Skeleton3D node and Godot maps the bones by name. The Rest Fixer options then match the model to the reference T-pose.
Does AI motion capture work for 2D games?
Rarely. Most 2D games use frame animation or cut-out rigs, so the useful AI tools are sprite-sheet generators such as Ludo.ai and PixelLab. Check generated frames for consistency before using them.